Neural-Network-Visualizer

Introduction

This project aims to visualize output of a basic 3 layered Neural Network(NN). (Made from scratch).

Neural networks are essentially universal function approximator, this program visualizes the output produced by a NN showing how, NN slowly estimates the provided function, while providing some customizable hyperparameters.

How to use

User can proveide a custom mathematical function y = f(x, z) (Note: automatic scaling is not provided for flexibility reason) or choose one from the already provided functions, then press Start for the neural network to start estimating it.

The gray figure is the estimated function which rapidly changes as the NN fits the function with lower loss meanwhile the colored figure is the desired output (which is visual representation of the function f(x, z) provided.

Interactivity

1. 3D camera is provided to better examine the results.

   • Left Click + Drag -> Rotate camera (about the clicked point)
   • Scroll -> to zoom in/out.

2. User can pause/resume the neural network to have a better look at its output.

3. Change learning rate to see how it affects the learning process.

4. Change iterations per epoch to speed up/slow down the learning process.

5. Change number of hidden nodes to see how it changes power of Neural Network to learn.